Transformers develop four algorithmic phases of in-context learning on Markov chains via two distinct multi-layer subcircuit mechanisms, with phase boundaries set by data diversity K.
To do this, we expand the loss to first order inβandδaround the unigram solution, L(β, δ) =L 1-Gen −c ββ−c δδ+
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Distinct mechanisms underlying in-context learning in transformers
Transformers develop four algorithmic phases of in-context learning on Markov chains via two distinct multi-layer subcircuit mechanisms, with phase boundaries set by data diversity K.